摘要:根据圆形图像目标中心对称特性，提出一种基于空间位置描述的抗旋转高效高鉴别二值模式特征提取方法.该方法在特征计算时，通过径向变换重新构建局部坐标系统，并在此基础上实现具有抗旋转变换的空间对称区域的局部二值模式提取；同时，池化操作采用了具有旋转不变性的环状空间划分，从而保证了最终特征描述的抗旋转变换能力.该方法分别在欧元硬币数据集、QQ表情数据集、车标数据集上进行了测试，识别准确率最高分别达到100%，100%，97.07%.在具有旋转情况的欧元硬币和QQ表情数据集中，优于传统的LBP特征和HOG特征，而且算法具有高效性，其单点计算时间只需0.045ms.



Abstract:According to the center symmetric characteristics of circular image objects, this study proposes an anti-rotation and efficient discriminative binary feature extraction method based on pairs of spatial symmetry structure. This method reconstructs the local coordinate system by radial transform during feature computation, based on it, then local binary pattern with anti-rotation of spatial symmetry regions are extracted. Meanwhile, the annular space is adopted to achieve rotation invariability during the feature pooling operation, which ensures the anti-rotation ability of final feature description. This method are tested in the euro coins, QQ expression, and car logo data set, and the recognition accuracy reached 100%, 100%, and 97.07% respectively, which is superior to traditional LBP and HOG features in euro coins and QQ expression datasets. Moreover, the algorithm is efficient, and the computation time for single point feature extraction is only 0.045 ms.



PDF全文下载地址:

http://jos.org.cn/jos/article/pdf/5566